Syndicate content

PhD INPhINIT “la Caixa” fellowship on Geometric analysis of video recordings

-Research Project:

Video content is ubiquitous nowadays thanks to the proliferation of digital cameras in the society and the constant publication and sharing of videos in different internet platforms such as YouTube, Vimeo, Twitter, etc. The purpose of this project is to study the dynamic content of videos in order to better understand the geometric 3D configuration of the scene that has produced the video. More specifically, our goal is to segment the scene in the different objects that compose it, infer the depth ordering of the objects, estimate their motion, trajectories and interactions and also complete the objects that are occluded. This constitutes a complete geometric analysis that may enrich the semantic understanding of the video. Different applications that may benefit from this analysis are automatic video postproduction, advanced driver assistance systems, 2D to 3D video conversion, and video summarization.

Research topics: optical flow, depth estimation, segmentation, inpainting, dynamic scene understanding.


-Job position description:

This research project will be supervised by Dr. Ballester and Dr. Haro, permanent faculty members of the Image Processing Group (IPG) at Universitat Pompeu Fabra, Barcelona. The research of the IPG lies at the intersection of image processing, computer vision, applied mathematics, computer science and technology applications.

The PhD position is offered to a highly motivated researcher, to join our team in image and video processing and computer vision. Prospective applicants should have a strong academic record with solid background in software development, and experience in computer vision or image processing. Good programming skills are expected, preferably in C/C++ and MATLAB/Python. Knowledge of deep-learning frameworks is a plus.

This research is related to the MINECO/FEDER project "Models and Computational Tecniques for the Analysis and Processing of Videos" with reference TIN2015-70410-C2-1-R and the european RISE project (Research and Innovation Staff Exchange) on Nonlocal Methods for Arbitrary Data Sources.


R. P.Palomares, E. Meinhardt-Llopis, C. Ballester, G. Haro. FALDOI: a new minimization strategy for large displacement variational optical flow. Journal of Mathematical Imaging and Vision, 58(1), 27-46, 2017.

P. Vitoria-Carrera, V. Fedorov, C. Ballester. Spatio-temporal tube segmentation through a video metrics-based patch similarity measure. IMVIP 2017.

M. Oliver, R. P.Palomares, C. Ballester, G. Haro. Spatio-temporal binary video inpainting via threshold dynamics. Proc. IEEE ICASSP, 2017.

B. Rezaeirowshan, C. Ballester, G. Haro. Monocular Depth Ordering using Perceptual Occlusion Cues. VISAPP, 2016.

M. Oliver, G. Haro, M. Dimiccoli, B. Mazin, C. Ballester. A Computational Model for Amodal Completion. Journal of Mathematical Imaging and Vision, 56(3), 511-534, 2016.

V. Fedorov, P. Arias, G. Facciolo, C. Ballester. Affine Invariant Self-Similarity for Exemplar-Based Inpainting. VISAPP 2016.